Key takeaways
- OpenAI announced Images 2.6 on November 4, 2026 across ChatGPT, Codex and the API. Its developer catalog now lists gpt-image-2.6-sunburst for detailed visual work and gpt-image-2.6-flare for fast iteration.
- Both API models support text and image input and image output on v1/images/generations and v1/images/edits. Their model pages currently mark v1/responses as unsupported, so builders should not assume the 2.5 Responses-tool integration has carried over.
- The launch adds low, medium and high complexity settings, markup-guided regional edits, static SVG output and animated GIF output. Complexity defaults to high and has no separate fee, but OpenAI says higher complexity can increase latency and unwanted changes.
- Sunburst and Flare have the same documented standard rates: $5 per million text-input tokens, $8 per million image-input tokens and $30 per million image-output tokens, with lower cached-input rates. Equal rates do not guarantee equal cost per image because token use can differ.
- Trial both variants on a frozen set of generation and editing tasks. Keep output-format constraints explicit, inspect text and brand consistency, record actual token use and latency, and retain the previous model route until the new one passes those gates.
OpenAI released Images 2.6 on November 4
OpenAI announced Images 2.6 on November 4, 2026, describing faster generation, GIF creation, more precise editing and new API controls across ChatGPT, Codex and the API. The API release has two exact identifiers: gpt-image-2.6-sunburst, positioned for detailed compositions and editing, and gpt-image-2.6-flare, positioned for interactive generation and fast iteration.
The model pages are live and list text and image input with image output. Both support POST /v1/images/generations and POST /v1/images/edits. OpenAI’s official RSS establishes the wider ChatGPT and Codex announcement, but AccessAllGPT did not sign in to confirm account-by-account rollout. The canonical announcement body was blocked by a Cloudflare challenge during review.
The API surface is narrower than the product announcement
For direct API use, the current model pages mark Image generation and Image edit as supported and mark Responses, Chat Completions, Realtime, Batch and fine-tuning as unsupported. That is an important migration detail: the general image guide still uses GPT Image 2.5 in its main Responses API examples, while its 2.6 section says the new output capabilities run through the Image API.
Treat that documentation state as the contract until OpenAI changes it. A ChatGPT or Codex feature being available does not imply the same orchestration path is available through v1/responses. Builders should pin the exact 2.6 model ID on the Images endpoint and add an endpoint-compatibility check before changing production traffic.
Complexity separates planning effort from render quality
Both models accept complexity values of low, medium or high; high is the default. OpenAI describes low complexity as suitable for focused edits and high complexity as potentially useful for dense text, complex layouts and relationships among objects. Quality remains a separate setting, with low, medium, high, xhigh and max documented for 2.6.
OpenAI says complexity has no separate fee, but that does not make higher settings free operationally: the guide warns that higher complexity can increase latency and introduce unwanted changes. Teams should therefore test complexity and quality as independent variables instead of setting both to their maximum and assuming the result is optimal.
Markup-guided edits give regional changes a clearer target
The edit endpoint can accept a separate transparent PNG annotation layer as markup. A user can circle or sketch over the region to change while keeping the original image separate, then describe the requested edit and preservation requirements in the prompt. OpenAI documents this for supported GPT Image 2.5 and 2.6 models.
The control is useful for product photos, layouts and brand assets, but it is not a deterministic mask. OpenAI tells developers to review the result, and the broader limitations still include precise composition, text placement and consistency across generations. Consequential publishing workflows should compare untouched regions as well as the intended edit.
SVG and GIF output arrive with strict constraints
For SVG, set output_format to svg on the generation endpoint and n to 1. The result is static vector content with no scripts, external resources or embedded raster images. SVG output does not support streaming, partial images, compression or prompt rewriting; the edit endpoint cannot emit SVG, and SVG is not accepted as a raster edit input.
For animated GIFs, set output_format to gif and describe the motion and loop. The API exposes 2–64 frames, a 20–2,000 millisecond frame duration and per-frame dimensions in multiples of 16. GIF generation does not support streaming, partial images, custom compression or prompt rewriting, and the edit endpoint cannot emit GIFs. Pixel budgets can force smaller frames, fewer frames or fewer animations.
Sunburst and Flare share token rates, not necessarily image costs
OpenAI lists the same standard rates for both models: $5 per million text-input tokens and $1.25 per million cached text-input tokens; $8 per million image-input tokens, $2 per million cached image-input tokens and $30 per million image-output tokens. Markup-guided edits use those rates without an additional markup fee. SVG also incurs authoring-model tokens in addition to the reference image.
Those are token rates, not a fixed price per image. OpenAI says equal rates do not imply equal image cost because models and quality settings can consume different token counts. A purchasing comparison therefore needs observed usage by size, quality, complexity, edit inputs and output format—not a single advertised rate.
Known limits still affect production design
OpenAI says complex prompts can take up to two minutes and documents remaining difficulty with exact text placement, recurring-character or brand consistency, and layout-sensitive composition. Rate limits vary by account. Some organizations may also need API organization verification, and OpenAI recommends a stable privacy-preserving safety_identifier for applications serving individual users.
The vendor documentation does not publish an independent quality benchmark, a guaranteed speed improvement, a fixed rollout percentage for ChatGPT or Codex, or a universal per-image cost. AccessAllGPT did not test either model. “Faster” and “more precise” remain OpenAI claims until measured on the buyer’s own workload.
What builders and buyers should do next
Trial Sunburst and Flare against the same frozen tasks: new images, local edits, text-heavy assets, repeated brand elements, SVG structure and GIF loop quality. Capture model ID, endpoint, size, quality, complexity, token use, wall-clock latency, moderation outcomes and human acceptance. Test low and medium complexity before accepting the high default.
Constrain the rollout to the Images endpoints and reviewed outputs while endpoint support remains narrower than GPT Image 2.5 examples. Wait if the product needs Responses API orchestration, deterministic layouts or a predictable per-image budget that has not been measured. Reject automatic migration when the new route cannot preserve brand assets, text, unedited regions or required accessibility and safety checks.
Copy-ready GPT Image 2.6 rollout record
Use one record for each product workflow and model variant so capability, cost and quality claims remain tied to an exact configuration.
Entries stay in this browser tab and are not submitted to AccessAllGPT. Blank responses are copied as [Unresolved].
Generation or edit; product image, illustration, layout, SVG or GIF; users, volume, owner and consequence of failure.
gpt-image-2.6-sunburst or gpt-image-2.6-flare, Images endpoint, SDK version, account tier, region and snapshot date.
Prompt template, input images, markup, size, quality, complexity, output format, frame controls, moderation and safety identifier.
Representative prompts and images, exact text, brand elements, recurring subjects, preservation regions, difficult layouts and blocked-content cases.
Acceptance rate, edit fidelity, untouched-region drift, text accuracy, consistency, SVG validity, GIF loop quality, latency, errors and retries.
Text/image input and output token use, cached use, observed cost per accepted asset, rejected generations, SVG authoring cost and monthly projection.
Human review, file validation, accessibility checks, provenance, content-policy handling, timeout, fallback model and rollback trigger.
Trial, constrain, wait or reject; chosen variant and settings, traffic ceiling, known exceptions, owner, expiry and retest date.
Primary sources
Browse the publication-wide evidence index →
- Introducing Images 2.6OpenAI · Reviewed: Official OpenAI News RSS title, summary, Product classification, canonical URL and November 4 publication timestamp; the canonical article returned a Cloudflare challenge during review · Retrieved · Supports: OpenAI announced Images 2.6 on November 4, 2026 and describes faster image generation, GIF creation, more precise editing and new API controls across ChatGPT, Codex and the API. The blocked article body was not used for additional claims.
- Image generation guideOpenAI Developer Documentation · Reviewed: GPT Image 2.6 overview; complexity controls; markup-guided edits; SVG and GIF output; output-format constraints; limitations; safety identifiers; cost and latency · Retrieved · Supports: OpenAI documents Sunburst and Flare, low/medium/high complexity, markup-guided edits, static SVG and animated GIF generation through the Image API, output constraints, known limitations and token-based cost behavior.
- GPT-Image-2.6 Sunburst model pageOpenAI Developer Documentation · Reviewed: Model identity, intended use, input and output modalities, text and image token prices, supported endpoints, snapshot alias and rate-limit note · Retrieved · Supports: The live model page identifies gpt-image-2.6-sunburst for detailed visual work, lists image generation and edit endpoints as supported, and lists the Images API token rates. Responses API support is marked unavailable on this model page.
- GPT-Image-2.6 Flare model pageOpenAI Developer Documentation · Reviewed: Model identity, intended use, input and output modalities, text and image token prices, supported endpoints, snapshot alias and rate-limit note · Retrieved · Supports: The live model page identifies gpt-image-2.6-flare for fast iteration, lists image generation and edit endpoints as supported, and lists the same Images API token rates as Sunburst. Responses API support is marked unavailable.
- OpenAI API pricingOpenAI Developer Documentation · Reviewed: Standard image-generation pricing table for GPT Image 2.6 Sunburst and Flare; text input, image input, cached input and image output token rates · Retrieved · Supports: OpenAI lists identical standard token rates for both GPT Image 2.6 variants: $5 per million text-input tokens, $1.25 cached; $8 per million image-input tokens, $2 cached; and $30 per million image-output tokens.
Limitations
This is a first-party launch and documentation review, not hands-on image-model testing or an independent benchmark. The canonical announcement body returned a Cloudflare challenge, so AccessAllGPT relied on OpenAI’s official RSS record for its date, summary and product-surface claims and on accessible developer documentation for API details. We did not sign in to ChatGPT or Codex; verify rollout by plan, account or geography; call gpt-image-2.6-sunburst or gpt-image-2.6-flare; generate, edit, stream or decode an image; inspect output provenance metadata; validate SVG structure or GIF animation; test latency, instruction following, text rendering, consistency, markup fidelity, safety filtering or rate limits; compare either model with GPT Image 2.5 or a competitor; or calculate a representative per-image price. OpenAI controls every cited surface, and its capability and speed descriptions are vendor claims rather than independent evidence.
Disclosures
AccessAllGPT received no model access, API credits, private demo, briefing, benchmark data, review, payment or compensation for this article. OpenAI did not sponsor, review or endorse it. AccessAllGPT did not test, score, rank or certify either image model. AccessAllGPT Research is operated by NeuralArc, is independent, and is not affiliated with OpenAI. Publication-wide relationships are listed on the disclosures page.
